For decades, Git has been the gold standard for version control. It's the bedrock of almost every modern software project. But at Anthropic, the AI lab behind Claude, they hit a wall—literally. As their AI agents began generating code at an inhuman pace, the very tools designed to manage human developers started to buckle.
The 30,000 Merge Meltdown
It happened fast. In just four months, the volume of code merges at Anthropic surged 12x. The culprit? Claude. As the company leveraged its own models to accelerate development, coding stopped being the bottleneck, but managing that code became a nightmare. When you're hitting 30,000 merges a day, a system built for human-paced commits starts to lag, glitch, and eventually break.
Enter Scion: Built by AI, for AI
Rather than waiting for a general-purpose fix from the industry, Anthropic took a radical step: they built their own version control system called Scion. In a poetic twist, a small team of seven engineers used Claude itself to help rebuild core infrastructure in just five months. Scion was designed specifically to handle the massive scale of a monorepo flooded with AI-generated changes, removing the friction that had turned Git into a productivity sink.
A Warning Shot for the Industry
Anthropic isn't releasing Scion as an open-source product; instead, they've shared their experience to signal to giants like GitHub and GitLab that the world is changing. We are entering an era where the speed of code production is no longer limited by how fast a human can type, but by how fast our infrastructure can merge.
As AI agents move from writing snippets to managing entire repositories, the industry may soon find that the tools of the 2000s simply aren't built for the scale of 2026.
Sources
Media
